Temporal Mining Mixture Model for Residential Occupancy Prediction

Conserving energy and optimizing its use has been a long standing challenge. Apart from the monetary benefits associated with successfully tackling these problems, saving energy has a significant positive environmental impact. One potentially fruitful approach would be to automatically adjust the HVAC of residential buildings based on occupancy. In this study, we mine people’s individual activity profiles to predict the occupancy of residential buildings. A novel hybrid method is proposed that uses episode mining for target event detection and a mixture of episode-generating HMM (EGH), combined with standard kNN approaches. We conclude by demonstrating how this hybrid approach consistently yields the best results.